Skip to main content
JAMA Network logoLink to JAMA Network
. 2026 Sep 21;83(11):1110–1120. doi: 10.1001/jamaneurol.2026.3253

A Harmonized Visual Reading Framework for Tau PET Staging in Alzheimer Disease

Emma Ruppert 1, Marie R Vermeiren 2,3,4, Marina Scop Medeiros 1, Guilherme Povala 1, Carolina Soares 1, Andreia Rocha 1, Alvaro de Oliveira Franco 1,5, Matheus Scarpatto Rodrigues 1, Markley Oliveira Jr 1, Rayan Mroué 1, Pamela C L Ferreira 1, Guilherme Bauer-Negrini 1, Firoza Z Lussier 1, Livia Amaral 1, Bruna Bellaver 1, Cécile Tissot 6,7, Joseph Masdeu 8, Dana L Tudorascu 1, Thomas Karikari 1, David N Soleimani-Meigooni 9, Juan Fortea 10,11, Val J Lowe 12, Hwamee Oh 13, Belen Pascual 8, Brian A Gordon 14, Pedro Rosa-Neto 7,15, Suzanne Baker 6, Rik Ossenkoppele 3,16, Elsmarieke van de Giessen 2,4, Tharick A Pascoal 1,✉, for the HEAD study
PMCID: PMC13595235  PMID: 42766299

Key Points

Question

Can a harmonized visual reading framework enable reliable, tracer-agnostic staging of tau positron emission tomography (PET) pathology for clinical use?

Findings

In this multicenter cross-sectional study of 681 participants scanned with multiple tau PET radiotracers, a 5-class harmonized visual reading framework demonstrated high interrater and intertracer agreement and identified more early-stage tau pathology than the regulatory-approved approach. Tau burden classes were associated with progressively worse cognition, greater amyloid-β burden, and higher plasma phosphorylated tau 217 (p-tau217) concentrations.

Meaning

A harmonized visual reading framework may enable consistent detection and staging of Alzheimer disease pathology, improve comparability across tracers, and support biomarker-informed clinical decision-making.

Abstract

Importance

Tau positron emission tomography (PET) can detect and stage Alzheimer disease pathology; however, current binary regulatory-approved visual interpretation focuses on advanced disease, limiting early detection and patient stratification.

Objective

To develop and validate a harmonized tau PET visual reading framework across radiotracers that stages tau burden into 5 classes: negative, low, moderate, high, and atypical.

Design, Setting, and Participants

This prospective multicenter cross-sectional study among a cohort with head-to-head tau PET acquisitions was conducted from March 2022 to August 2025 and data analyzed from February to June 2026, with blinded visual reads performed by clinicians. The study was conducted at 9 sites across North America and Europe. Of those individuals enrolled in the HEAD (Head-to-Head Harmonization of Tau Tracers in Alzheimer's Disease) study, a proportion of participants aged 50 to 90 years who completed multitracer tau PET (2-4 scans) and cognitive assessments were included, spanning cognitively unimpaired to dementia.

Exposures

Tau PET with 18F-MK-6240 (TAUKLARIFY, n = 681), 18F-flortaucipir (Tauvid, n = 644), 18F-RO948 (n = 150), and 18F-PI-2620 (n = 149); all participants underwent amyloid-β (Aβ) PET and magnetic resonance imaging.

Main Outcomes and Measures

The primary outcomes were interrater and intertracer agreement for 5-class tau classification; prevalence ratio (PR) vs the regulatory-approved 18F-flortaucipir framework; and associations with cognition, Aβ burden, and plasma phosphorylated tau 217 (p-tau217).

Results

Of 822 individuals enrolled in the HEAD study, 681 participants spanning cognitively unimpaired to dementia were included. Mean (SD) age was 69.3 (8.4) years, and 367 participants (53.9%) were female. The harmonized 5-class framework achieved excellent interrater agreement (κ = 0.77-0.85) and good to excellent intertracer agreement, highest between 18F-MK-6240 and 18F-flortaucipir (κ = 0.88). Compared with the regulatory-approved framework, the harmonized classification identified more tau-positive cognitively unimpaired participants (regulatory-approved binary: 26 of 364 participants; harmonized 18F-flortaucipir: 49 of 364 [PR = 1.89; 95% CI, 1.31-2.70; P = .001]; harmonized 18F-MK-6240: 71 of 364 [PR = 2.73; 95% CI, 1.95-3.83; P < .001]) and those with mild cognitive impairment (regulatory-approved binary: 100 of 223; harmonized 18F-flortaucipir: 123 of 223 [PR = 1.23; 95% CI, 1.12-1.35; P < .001]; harmonized 18F-MK-6240: 142 of 223 [PR = 1.42; 95% CI, 1.27-1.59; P < .001]), with convergence in dementia. Tau burden classes showed stepwise worsening of cognition, Aβ PET burden, and plasma p-tau217, with 1.4-fold, 3.2-fold, and 2.5-fold differences between the low and high tau classes, respectively.

Conclusions and Relevance

In this prospective multitracer cross-sectional study, a 5-class harmonized tau PET visual reading framework showed high interrater and intertracer agreement and detected more early-stage tau pathology than the regulatory-approved binary approach, with tau burden classes tracking stepwise with cognition and biomarkers. These findings support a standardized, tracer-agnostic framework for diagnostic categorization, therapeutic window identification, and clinical trial stratification.


This cross-sectional study develops and validates a harmonized tau positron emission tomography (PET) visual reading framework across radiotracers that stages tau burden into 5 classes.

Introduction

Amyloid-β (Aβ) plaques and tau neurofibrillary tangles are the defining neuropathological hallmarks of Alzheimer disease (AD).1 Both can be detected in vivo using positron emission tomography (PET), reflecting neuropathological stages.2,3,4,5,6,7,8 Currently, Aβ PET visual assessment supports diagnostic classification and eligibility for anti-Aβ therapies.9,10 However, although Aβ deposition is key to confirming the presence of AD pathology, it has a weak association with imminent cognitive decline.2,11 Indeed, in an Aβ-positive individual with cognitive decline, a negative tau PET scan raises the possibility that non-AD etiology may be the primary driver of symptoms.3 Tau PET, in contrast, is strongly associated with cognitive impairment and clinical presentation12,13,14 and can accurately distinguish AD from other dementias,8,15,16,17,18 positioning it to guide diagnosis, prognosis, and selection for disease-modifying therapies. Tau PET’s clinical utility depends on how scans are interpreted, yet the current regulatory-approved visual framework prioritizes specificity for advanced tau pathology, limiting the identification of early tau pathology.10

As tau PET expands into clinical settings, feasible and reliable visual reading methods are essential.19,20,21,22,23,24,25 Despite recent US Food and Drug Administration (FDA) approval of MK-6240 (TAUKLARIFY [Lantheus]; 18F-florquinitau),26 flortaucipir (Tauvid [Lilly]; previously 18F-T807 or 18F-AV1451) remains the only tau PET radiotracer with FDA- and European Medicines Agency–approved visual reading methods for adults with cognitive impairment (CI) undergoing assessment for AD.14,25,27,28 Tau PET’s principal advantage over fluid biomarkers is its ability to enable spatially resolved detection and regional staging of tau pathology. However, existing assessment methods do not fully exploit this capability.29 The current regulatory-approved framework, optimized for specificity, classifies tau PET as binary positive or negative, corresponding primarily to advanced pathology (Braak V and VI).25,30,31 This approach has limited sensitivity for detecting earlier tau deposition (Braak I-IV), the stage where patients are most likely to benefit from therapeutic intervention.3,32 This classification also does not capture the extent of tau accumulation, limiting differentiation between moderate and widespread tau deposition, despite strong associations between tau distribution and CI and expert agreement that tau PET should be used for staging in practice.25 Furthermore, this method was developed and validated exclusively for 18F-flortaucipir, limiting cross-study comparability and broader clinical adoption.

To address these gaps, we developed a harmonized visual reading framework for tau PET staging across 18F-flortaucipir, 18F-MK-6240, 18F-RO948, and 18F-PI-2620 in a large multisite cohort. The framework staged tau PET burden into 5 classes—negative, low, moderate, high, and atypical—designed to capture early tau deposition. We compared the harmonized framework with the regulatory-approved tracer-specific approach and evaluated its biological and clinical validity across the AD continuum.

Methods

The study protocol was approved by all participating institutional review boards, and all participants provided written informed consent by self or proxy. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Participants

This cross-sectional analysis used baseline data from a portion of participants enrolled in the HEAD (Head-to-Head Harmonization of Tau Tracers in Alzheimer's Disease) study, an ongoing longitudinal multicenter cohort (ClinicalTrials.gov identifier: NCT05361382). Inclusion and exclusion criteria are in eFigure 1 in Supplement 1. Participants aged 50 to 90 years were enrolled from March 2022 to August 2025 across 9 sites in North America and Europe and underwent cognitive assessments using the National Alzheimer's Coordinating Center Uniform Data Set version 3 (UDS-3), supplemented by the Mini-Mental State Examination. Cognitively unimpaired (CU) individuals had normal cognition and Clinical Dementia Rating (CDR) of 0. Participants with mild CI (MCI) had CDR of 0.5; and participants with dementia had CDR of 1.0 or more.33

Imaging

eMethods 1 in Supplement 1 details magnetic resonance imaging (MRI) and PET acquisition and processing. Aβ PET used site-specific radiotracers (eTable 1 in Supplement 1). Aβ standardized uptake value ratios (SUVRs) were calculated using whole cerebellar gray matter as the reference region and converted to Centiloids.34 Two pairs of raters independently determined Aβ positivity, with a tiebreaker resolving disagreements. Tau PET imaging with 18F-MK-6240 was acquired across all sites, while 18F-flortaucipir, 18F-PI-2620, and 18F-RO948 were each acquired at select sites (eTable 2 in Supplement 1). Tau SUVRs were calculated using inferior cerebellar gray matter as the reference region, and mean SUVRs were extracted from regions of interest (ROIs) using the Cerebrum Atlas.35 Blood sample measurements are detailed in eMethods 2 in Supplement 1.

Harmonized Visual Reading Method

To develop the harmonized visual reading method, we defined voxel-level abnormality as more than 2 SDs above the mean of CU Aβ-negative individuals (n = 51) scanned with all 4 tau PET radiotracers (n = 131) (eTable 3 in Supplement 1), generating individual maps in Montreal Neurological Institute (MNI) 152 space. For each ROI, the proportion of participants with abnormal uptake was calculated by averaging individual maps. Then, ROIs were grouped into 7 composite regions based on frequency of abnormality across radiotracers and anatomical proximity (eFigure 2 in Supplement 1), providing a data-driven approximation of the hierarchical progression of tau pathology. Additional thresholds had no impact on composite assignments (eFigure 3, eTable 4 in the Supplement). Regions susceptible to off-target or spillover binding were excluded for all radiotracers (eFigure 4 in Supplement 1).36,37,38,39

We then developed the following 5-class classification scheme based on the pattern of composite region involvement: negative, low, moderate, high, and atypical tau burden (Figure 1). For the harmonized framework, binary tau positivity corresponds to any classification other than negative. This framework reflects established spatial progression patterns of tau pathology in AD31 and enables both early detection and disease staging. To standardize visual assessment, we developed a web-based visual reading platform that presents tau PET images coregistered with MRI and guides raters through binary classification (positive or negative) of each composite. To aid raters, the platform overlaid composite-specific masks (refined using 3D Slicer40) (eFigure 5 in Supplement 1). This standardized interface ensures consistency across raters and reduces variability associated with different software platforms (eFigure 6 in Supplement 1). On completion of all composite ratings, raters assigned a confidence level from 1 to 5 and could add commentary. The platform then assigns the corresponding 5-class classification. Three clinician-scientist raters (E.R., M.R.V., E.v.d.G.) with neuroimaging experience and 3, 2, and 5 years of tau PET reading experience, respectively, evaluated all scans following standardized instructions (eMethods 3 in Supplement 1). Raters were blinded to demographic, clinical, and biomarker data and other raters’ assessments. Washout periods between reading sessions minimized recall bias for participants with multiple scans. For both binary positivity and 5-class classification, majority vote was applied, with discordant cases across all raters resolved by consensus. Three clinician-scientist raters (E.R., M.S.M., A.d.O.F.) also assessed 18F-flortaucipir images blinded using the regulatory-approved visual reading framework, with majority vote applied.14,25

Figure 1. Visualization of Brain Regions, Classification Schematic, and Reading Tool Showing Harmonized Framework for Tau Positron Emission Tomography (PET) Visual Reading and Classification.

Anatomical diagram, table, and tau P E T reading tool in three panels. Three labeled panels arranged vertically. Panel A, title text at top left: Graphic representation of masks used to assist the visual reads. Four 3 D brain surface renderings in a row with view labels beneath: Superior, Inferior, Lateral, Medial. Cortical and subcortical regions are overlaid with seven distinct solid colors. A legend at upper right labeled Composite with colored squares and numbers 1 through 7; colors include two blues for 1 and 2, three purple shades for 3, 4, and 5, and two greens for 6 and 7. Panel B, title text: Harmonized tau P E T visual reads classification criteria. A table with six columns and four rows, including the header row. Column headers from left to right: a blank row label column, Negative, Low, Moderate, High, Atypical. Row labels in the first column: Composite 1 and or 2; Composite 3, 4, and or 5; Composite 6 and or 7. Cell entries use symbols: minus, plus, and plus slash minus. Row Composite 1 and or 2: Negative minus; Low plus; Moderate plus; High plus; Atypical minus. Row Composite 3, 4, and or 5: Negative minus; Low minus; Moderate plus; High plus; Atypical plus slash minus. Row Composite 6 and or 7: Negative minus; Low minus; Moderate minus; High plus; Atypical plus slash minus. Panel C, title text: Harmonized tau P E T visual reading tool. Three brain images labeled across the top: Sagittal, Coronal, Axial, each with a rainbow heatmap overlay and a color bar beneath ranging from zero point three zero to two point three zero. Lower left text: Value: zero point seven one; Region: zero. Lower right text: Slice: 94, 114, 94. Along the bottom, user interface controls with labels including Colormap, Opacity, Invert, Smoothing, Min, Max, Contrast, Brightness, Gamma, Show, Mask, and Outline, with sliders and checkboxes.

A, Seven composite regions used for visual assessment, displayed in superior, inferior, lateral, and medial views. The composite cortical regions were (1) amygdala and entorhinal; (2) fusiform gyrus; (3) middle and inferior temporal; (4) posterior cingulate and precuneus; (5) inferior parietal, lateral occipital, superior temporal, superior parietal, and supramarginal; (6) anterior cingulate, middle frontal, superior frontal, inferior frontal, and orbitofrontal; and (7) precentral, paracentral, and postcentral. Composites are color coded by stage: early (composites 1 and 2, in blue), moderate (composites 3, 4, and 5, in purple), and late (composites 6 and 7, in green). B, Stepwise classification algorithm for tau burden staging. Scans without abnormal tracer uptake in any composite are classified as tau PET negative. Scans with abnormal uptake are classified as tau PET positive and further categorized as low (uptake limited to composites 1 and/or 2), moderate (low criteria met plus uptake in at least 1 of the composites 3, 4, and/or 5), high (moderate criteria met plus uptake in at least 1 of the composites 6 and/or 7), or atypical (uptake in any of the composites 3-7 without involvement of composites 1 or 2). Panel C shows a freely available web-based tool (https://visualreading.unitau.app/) that visually guides raters through the 7 composite regions. The tool processes all data locally on the user’s computer, and data are not transferred to external servers.

Statistical Analysis

Interrater agreement was quantified using unweighted Cohen κ for pairwise comparisons and Fleiss κ for 3-rater comparisons. Intertracer agreement was quantified using unweighted Cohen κ for binary positivity and weighted Cohen κ for 5-class classification. Kappas were interpreted as poor (<0.40), fair (0.40-0.59), good (0.60-0.74), and excellent (≥0.75).41 To assess robustness, sensitivity analyses were performed restricted to CI and Aβ-positive subgroups, participants with all 4 radiotracers (n = 131), participants excluded from the initial composite-development sample, and participant subgroups stratified by scan interval. Participants with tau PET data available for fewer than 4 radiotracers were included in all analyses for which they had data.

Comparisons between the harmonized and regulatory-approved frameworks were restricted to participants with both 18F-flortaucipir and 18F-MK-6240 tau PET (n = 644). To compare tau PET positivity prevalence between frameworks, generalized estimating equations with a Poisson distribution, log link, and exchangeable within-person correlation structure were fitted with Centiloid as a covariate to estimate prevalence ratios (PRs). Differences across clinical groups were evaluated using Wald interaction tests and estimated stratum-specific PRs with 95% confidence intervals using marginal means. To estimate positivity as a function of age, we fit generalized additive models with a binomial link function and age modeled as a cubic regression spline with knots at 60, 70, and 80 years. To compare Montreal Cognitive Assessment (MoCA) scores, Centiloid values, and plasma phosphorylated tau 217 (p-tau217) across tau burden classes, linear models adjusted for age and sex were fitted separately for 18F-flortaucipir and 18F-MK-6240. Pairwise contrasts were estimated using marginal means with Tukey correction for multiple comparisons. Domain-specific cognitive composites were derived from the UDS-3 neuropsychological battery and z-scored for age, sex, and education.42 Missing data were minimal (Table). Participants missing outcome data were excluded only from analyses involving those measures. All tests were 2-sided with a significance threshold of P < .05, and all analyses were conducted in R version 4.4.0 (R Foundation).

Table. Demographics and Key Characteristics of HEAD Study Participants (n = 681).

Characteristic No. (%)a
CU (n = 369) MCI (n = 245) Dementia (n = 67)
Age, mean (SD), y 67.2 (8.3) 71.8 (7.7) 72.0 (8.7)
Sex
Female 224 (60.7) 112 (45.7) 31 (46.3)
Male 145 (39.3) 133 (54.3) 36 (53.7)
Education, mean (SD), y 16.1 (2.5) 15.5 (3.4) 15.5 (2.9)
Raceb
Asian 4 (1.1) 7 (2.9) 0
Black or African American 15 (4.1) 5 (2.0) 3 (4.5)
White 348 (94.3) 231 (94.3) 63 (94.0)
Other or unknownc 2 (0.5) 2 (0.8) 1 (1.5)
Ethnicityb
Hispanic or Latino 7 (1.9) 13 (5.3) 6 (9.0)
Non-Hispanic or Latino 362 (98.1) 232 (94.7) 61 (91.0)
APOE ε4
Carrier 114 (31.2) 123 (50.2) 42 (63.6)
Noncarrier 251 (68.8) 122 (49.8) 24 (36.4)
MMSE score, mean (SD) 28.7 (1.5) 26.4 (3.1) 20.9 (6.1)
MoCA score, mean (SD) 27.1 (2.2) 22.7 (4.2) 15.8 (6.5)
Aβ PET visual read
Negative 283 (76.7) 90 (36.7) 11 (16.4)
Positive 86 (23.3) 155 (63.3) 56 (83.6)
Centiloid, mean (SD) 14.5 (32.4) 57.6 (55.1) 90.6 (56.3)
Plasma p-tau217, mean (SD), pg/mL 0.4 (0.3) 0.8 (0.6) 1.4 (0.8)
Tau PET images
18F-MK-6240 369 (100) 245 (100) 67 (100)
18F-Flortaucipir 364 (98.6) 223 (91.0) 57 (85.1)
18F-RO948 71 (19.2) 60 (24.5) 19 (28.4)
18F-PI-2620 71 (19.2) 61 (24.9) 17 (25.4)

Abbreviations: Aβ, amyloid-β; APOE ε4, apolipoprotein ε4 allele; CU, cognitively unimpaired; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment; p-tau217, phosphorylated tau 217; PET, positron emission tomography.

a

Missing values: education (n = 4), APOE ε4 (n = 5), MMSE score (n = 2), MoCA score (n = 3), and plasma p-tau217 (n = 13).

b

Self-reported following the Uniform Data Set version 3 A1 package.

c

Included individuals identifying as Ashkenazi Jewish, Filipino, Latin-American, and unknown.

Results

Of 822 consented individuals, 74 withdrew, 10 had incomplete data, and 57 were excluded (eFigure 1 in Supplement 1). Of 681 participants aged 50 to 90 years, mean (SD) age was 69.3 (8.4) years, and 367 participants (53.9%) were female. 18F-MK-6240 was available in 681 participants (100%), 18F-flortaucipir in 644 (94.6%), 18F-RO948 in 150 (22.0%), and 18F-PI-2620 in 149 (21.9%). Mean (SD) interval between PET scans was 41.3 (46.6) days. Cohort characteristics are summarized in the Table and in eTables 5 through 8 in Supplement 1.

Agreement Rates

Interrater agreement for tau PET binary positivity and 5-class classification was excellent across radiotracers (κ = 0.77-0.93) (Figure 2A). Binary and 5-class agreement ranged from good to excellent in Aβ-positive, CI, and 4-tracer subsamples (eTable 9 in Supplement 1). Mean SUVR of tau distribution across classification groups are provided in eFigures 7 through 11 in Supplement 1. Composite-level agreement and rater confidence were notably lower for 18F-PI-2620, particularly for composite 1 (eFigures 12 and 13 in Supplement 1). Exclusion of composite 1 for 18F-PI-2620 improved interrater agreement while largely preserving intertracer agreements (eTable 10 and 11 in Supplement 1).

Figure 2. Bar Graphs Showing Interrater and Intertracer Agreement for Harmonized Tau Positron Emission Tomography Visual Reading.

Two-panel bar charts of agreement percent by tracer and tracer pairs. Two side-by-side panels labeled A and B, each a grouped bar chart with vertical axis labeled Agreement, percent, ranging from zero to 100 with tick marks at 20-unit intervals. Panel A title at upper left: Interrater agreement. Horizontal axis lists four categories: M K dash 6240, Flortaucipir, R O 948, and P I dash 2620. Each category contains two adjacent bars: a blue-gray bar for Binary and an orange bar for 5-Class classification. Approximate bar heights in panel A: M K dash 6240 about 91 percent (Binary) and about 84 percent (5-Class); Flortaucipir about 94 percent and about 86 percent; R O 948 about 85 percent and about 81 percent; P I dash 2620 about 97 percent and about 92 percent. Panel B title: Intertracer agreement. A legend box in the upper right of panel B maps blue-gray to Binary and orange to 5-Class classification. Horizontal axis in panel B lists six tracer pairs, each printed on two lines: M K dash 6240 with Flortaucipir; M K dash 6240 with R O 948; M K dash 6240 with P I dash 2620; Flortaucipir with R O 948; Flortaucipir with P I dash 2620; R O 948 with P I dash 2620. Each pair has two adjacent bars in the same colors as panel A. Approximate bar heights in panel B: M K dash 6240 and Flortaucipir about 89 percent and about 82 percent; M K dash 6240 and R O 948 about 88 percent and about 84 percent; M K dash 6240 and P I dash 2620 about 81 percent and about 75 percent; Flortaucipir and R O 948 about 86 percent and about 83 percent; Flortaucipir and P I dash 2620 about 85 percent and about 80 percent; R O 948 and P I dash 2620 about 88 percent and about 83 percent. No error bars or data labels appear on the bars.

A, Interrater agreement for each radiotracer. Grouped bars show percentage agreement for binary positivity and 5-class tau burden classification based on agreement across all 3 raters. Sample sizes were: 18F-MK-6240 (n = 681), 18F-flortaucipir (n = 644), 18F-RO948 (n = 150), and 18F-PI-2620 (n = 149). B, Intertracer agreement for each radiotracer pair. Grouped bars show percentage agreement for binary positivity and 5-class framework based on agreement across all 3 raters. Sample sizes were: 18F-MK-6240/18F-flortaucipir (n = 644), 18F-MK-6240/18F-RO948 (n = 150), 18F-MK-6240/18F-PI-2620 (n = 149), and all other pairs (n = 131 each).

Intertracer agreement (Figure 2B) was highest between 18F-MK-6240 and 18F-flortaucipir for both binary positivity (κ = 0.76) and 5 classes (κ = 0.88). Across all tracer pairs, binary agreement ranged from 81.2% to 88.5% and 5-class agreement ranged from 75.2% to 83.3%, with comparable ranges across Aβ-positive, CI, and 4-tracer subsamples and across scan intervals (eTable 12 and 13 in Supplement 1). Disagreements across radiotracers showed notable asymmetries (eFigure 14 in Supplement 1): 18F-MK-6240 detected positivity more frequently than other tracers (eFigure 15 and 16 in Supplement 1), while 18F-PI-2620 was more often negative when other tracers were positive, consistent with the removal of composite 1 (eTable 11-14 in Supplement 1). There was no convergence of atypical classification across tau PET tracers. Among participants with an atypical classification (n = 15) (eFigure 17, eTable 15 in Supplement 1), 10 of 11 Aβ-positive individuals (90%) showed a typical pattern of tau positivity on all other tracers. In contrast, all Aβ-negative individuals were tau-negative on all other tracers.

Tau PET Positivity and Staging

Binary tau PET positivity increased progressively from CU to MCI to dementia across radiotracers, with rates highest for 18F-MK-6240 and lowest for 18F-PI-2620 (eTable 16 in Supplement 1). Five-class classification showed that higher tau burden classes were more prevalent with greater CI and Aβ positivity across all radiotracers (eFigure 18 in Supplement 1). CU participants constituted most of the negative classification (18F-MK-6240: 296 [74.7%]; 18F-flortaucipir: 315 [74.1%]; 18F-RO948: 67 [69.1%]; 18F-PI-2620: 70 [58.8%]) (eFigure 19 in Supplement 1). Among the positive stages, the proportion of participants with CI increased progressively from low to high, with high tau burden composed almost exclusively of individuals with CI (18F-MK-6240: 110 [94.8%]; 18F-flortaucipir: 88 [95.7%]; 18F-RO948: 22 [100%]; and 18F-PI-2620: 17 [100%]).

Tau positivity increased with age for both 18F-MK-6240 and 18F-flortaucipir, regardless of visual reading framework (Figure 3). Harmonized reads identified more tau-positive individuals than the regulatory-approved framework, particularly in earlier stages (eTable 17 in Supplement 1). After Centiloid adjustment, harmonized 18F-flortaucipir was more likely to classify individuals as tau-positive than the regulatory-approved framework (219 of 644 [34.0%] vs 171 of 644 [26.6%]; PR = 1.28; 95% CI, 1.18-1.40), as was harmonized 18F-MK-6240 (263 of 644 [40.8%] vs 171 of 644 [26.6%]; PR = 1.54; 95% CI, 1.40-1.69). This effect varied by clinical class, with the largest divergence in CU (regulatory-approved 18F-flortaucipir, 26 of 364; harmonized 18F-flortaucipir, 49 of 364 [PR = 1.89; 95% CI, 1.31-2.70; P = .001]; harmonized 18F-MK-6240, 71 of 364 [PR = 2.73; 95% CI, 1.95-3.83; P < .001]) and MCI (100 of 223; 123 of 223 [PR = 1.23; 95% CI, 1.12-1.35; P < .001] and 142 of 223 [PR = 1.42; 95% CI, 1.27-1.59; P < .001]; respectively). In dementia, positivity rates converged (45 of 57, 47 of 57, and 50 of 57, respectively).

Figure 3. Bar and Line Graphs Showing Tau Positron Emission Tomography (PET) Positivity by Age and Clinical Diagnosis Across Visual Reading Frameworks.

Four-panel figure of tau positivity frequency by age and prevalence ratios. Four panels labeled A through D arranged in a two by two grid. Panel A title: M K dash 6 2 4 0 harmonized visual reading. Vertical axis labeled Frequency, percent, from 0 to 100 with gridlines; horizontal axis labeled Age, y, with bins 50 to 55, greater than 55 to 60, greater than 60 to 65, greater than 65 to 70, greater than 70 to 75, greater than 75 to 80, and greater than or equal to 80. Stacked bars at each age bin with legend Tau burden: Atypical in orange, Low in light blue, Moderate in medium blue, High in dark blue. A black smooth curve overlays the bars, rising from about 10 percent near age 50 to 55 to about 75 percent near age greater than or equal to 80. Total stacked bar heights increase with age, approximately 13, 22, 21, 34, 46, 60, and 72 percent across bins; the dark blue segment forms the base in each bar, with larger light blue segments in older bins and small orange caps only in the oldest bins. Panel B title: Flortaucipir harmonized visual reading. Same axes labels and age bins; same Tau burden legend and stacked color scheme. A black smooth curve rises from about 10 percent at 50 to 55 to about 62 percent at greater than or equal to 80. Total bar heights increase across age bins, approximately 13, 15, 20, 32, 39, 48, and 55 percent, with small orange atypical caps appearing from about greater than 60 to 65 onward. Panel C title: Flortaucipir regulatory dash visual reading. Same axes labels and age bins; legend with a beige bar labeled Positive tau and a black line labeled Estimated positivity frequency. Beige bars rise with age, approximately 13, 13, 15, 24, 28, 41, and 43 percent; the black curve rises from about 9 percent at 50 to 55 to about 42 percent at greater than or equal to 80. Panel D title: Tau positivity prevalence by visual reading frameworks across diagnoses. Left side contains a table-like listing under headers Cognitive status, Prevalence ratio with 95 percent C I, and Regulatory-approved flortaucipir. Rows grouped as C U, M C I, and Dementia; each group lists Harmonized M K dash 6 2 4 0 and Harmonized flortaucipir with printed prevalence ratios and confidence intervals: for C U, 2 point 7 3 with 1 point 9 5 to 3 point 8 3, and 1 point 8 9 with 1 point 3 1 to 2 point 7 0; for M C I, 1 point 4 2 with 1 point 2 7 to 1 point 5 9, and 1 point 2 3 with 1 point 1 2 to 1 point 3 5; for Dementia, 1 point 1 1 with 1 point 0 1 to 1 point 2 2, and 1 point 0 4 with 0 point 9 8 to 1 point 1 1. To the right, a forest plot on a horizontal axis labeled Prevalence ratio with 95 percent C I, spanning 0 to 4, includes a vertical dotted reference line at 1. Six square markers with horizontal confidence interval lines align to the listed rows, with markers positioned near 2 point 7 3, 1 point 8 9, 1 point 4 2, 1 point 2 3, 1 point 1 1, and 1 point 0 4.

Stacked bars show the proportion of tau-positive participants within 5-year age bins, stratified by tau burden class (low, moderate, high, and atypical) for the harmonized framework with 18F-MK-6240 (A) and 18F-flortaucipir (B) and by binary classification for the regulatory-approved framework with 18F-flortaucipir (C). Superimposed curves show estimated positivity frequency as a function of age, derived from generalized additive models. D, Prevalence ratios (PRs) comparing tau PET positivity for harmonized 18F-MK-6240 and harmonized 18F-flortaucipir reads relative to regulatory-approved 18F-flortaucipir reads, stratified by clinical diagnosis. Points represent stratum-specific PRs adjusted for Centiloid values with 95% confidence intervals. PR >1 indicates that the harmonized framework classified more participants as tau positive than the regulatory-approved framework. For harmonized 18F-flortaucipir vs regulatory-approved: cognitively unimpaired (CU), PR = 1.89 (95% CI, 1.31-2.70; P = .001); mild cognitive impairment (MCI), PR = 1.23 (95% CI, 1.12-1.35; P < .001); and dementia, PR = 1.04 (95% CI, 0.98-1.11; P = .16). For harmonized 18F-MK-6240 vs regulatory-approved: CU, PR = 2.73 (95% CI, 1.95-3.83; P < .001); MCI, PR = 1.42 (95% CI, 1.27-1.59; P < .001); and dementia, PR = 1.11 (95% CI, 1.01-1.22; P = .03). Analyses were restricted to participants with both 18F-MK-6240 and 18F-flortaucipir scans (n = 644).

Stepwise Association With CI, Cognition, and Biomarkers

Among 644 participants with 18F-MK-6240 and 18F-flortaucipir, higher tau burden classes in the harmonized framework were associated with progressively greater proportions of CI, poorer cognitive performance, higher Aβ load, and higher plasma p-tau217 (Figure 4). The prevalence of CI was lowest in the negative class for both radiotracers and increased stepwise across subsequent classes. MoCA scores did not differ between negative and low tau burden groups for 18F-MK-6240 (adjusted mean difference, 0.7; 95% CI, −0.4 to 1.9) or 18F-flortaucipir (adjusted mean difference, 0.5; 95% CI, −0.9 to 1.9). In subsequent classes, MoCA scores declined progressively: with 18F-MK-6240, by 2.5 points (95% CI, 0.9-4.1) from low to moderate and 4.6 points (95% CI, 3.0-6.2) from moderate to high; and with 18F-flortaucipir, by 3.3 points (95% CI, 1.6-5.0) and 4.4 points (95% CI, 2.8-6.0), respectively. Domain-specific cognitive composites showed a similar stepwise pattern (eTables 18 and 19 in Supplement 1).

Figure 4. Bar Graphs and Box Plots Showing Clinical Characterization of Tau Burden Classes.

Eight-panel figure: bars and box plots across tau burden classes for two tracers. Two-row, four-column chart with eight panels. Top row labeled A at far left with title text M K hyphen 6 2 4 0; bottom row labeled B at far left with title text Flortaucipir. In each row, the leftmost panel is a vertical bar chart with vertical axis labeled Cognitively impaired, percent, from 0 to 100, and horizontal categories Negative, Low, Moderate, High. Bar heights in row A are approximately 23 percent, 49 percent, 71 percent, and 92 percent, respectively; in row B approximately 26 percent, 43 percent, 86 percent, and 94 percent. The remaining three panels per row are box plots with overlaid individual points, using the same four x-axis categories and consistent category colors: Negative in light blue, Low in orange, Moderate in tan, High in gray. Second column panels: vertical axis labeled MoCA score, from 0 to 30. Medians decrease across classes, with row A medians near 27, 26, 24, and 18; row B near 26, 25, 22, and 18. Above these plots, horizontal comparison brackets include one labeled N S between Negative and Low, and multiple brackets labeled a spanning other group comparisons. Third column panels: vertical axis labeled Centiloids, from 0 to 250. Medians increase across classes; row A medians near 0, 15, 80, and 120; row B near 0, 15, 90, and 120. Several comparison brackets above include labels a, and in row B one bracket labeled b near the left. Rightmost panels: vertical axis labeled Plasma p tau 2 1 7, from 0 to 4. Medians increase across classes; row A medians near 0 point 25, 0 point 5, 0 point 85, and 1 point 3; row B near 0 point 35, 0 point 6, 1 point 0, and 1 point 3. Multiple comparison brackets above are labeled a, and in row B one bracket labeled b near the left. No legend box present.

Bar charts show the percentage of cognitively impaired participants (mild cognitive impairment or dementia) within each tau burden class (negative, low, moderate, and high) for 18F-MK-6240 (A) and 18F-flortaucipir (B). Box plots show Montreal Cognitive Assessment (MoCA) scores, Centiloid values, and plasma phosphorylated tau 217 (p-tau217) levels stratified by harmonized tau burden class for each radiotracer. Participants classified as atypical were excluded due to small sample size (18F-MK-6240, n = 1; 18F-flortaucipir, n = 7). Boxes represent the interquartile range, horizontal lines indicate the median, whiskers extend to 1.5 × the interquartile range, and individual data points are shown. Pairwise comparisons were estimated using linear models adjusted for age and sex, with Tukey correction for multiple comparisons. NS indicates not significant.

aSignificant at P < .001.

bSignificant at P < .01.

Centiloid values (CL) increased across tau burden classes with both radiotracers. With 18F-MK-6240, adjacent-class differences were 24.4 CL (95% CI, 14.7-34.1), 35.3 CL (95% CI, 21.4-49.2), and 48.1 CL (95% CI, 34.4-61.9); and with 18F-flortaucipir, 17.2 CL (95% CI, 4.9-29.6), 53.0 CL (95% CI, 37.3-68.7), and 32.8 CL (95% CI, 18.6-47.0). Plasma p-tau217 followed the same pattern. With 18F-MK-6240, adjacent-class differences were 0.2 pg/mL (95% CI, 0.07-0.29) from negative to low, 0.3 pg/mL (95% CI, 0.12-0.44) from low to moderate, and 0.6 pg/mL (95% CI, 0.46-0.77) from moderate to high. For 18F-flortaucipir, the corresponding differences were 0.2 pg/mL (95% CI, 0.04-0.32), 0.4 pg/mL (95% CI, 0.24-0.59), and 0.4 pg/mL (95% CI, 0.29-0.61). Across visual reading frameworks, tau-positive participants had lower MoCA scores and higher Aβ burden and plasma p-tau217 levels than tau-negative participants (eFigure 20 in Supplement 1). Participants classified as tau positive only by the harmonized framework had significantly higher CLs and plasma p-tau217 levels than those negative on both frameworks, whereas participants positive only on the regulatory-approved framework did not show consistent biomarker elevation (eFigure 21 in Supplement 1).

Discussion

We developed and validated a tracer-agnostic, harmonized visual reading framework for tau PET applicable across 18F-MK-6240, 18F-flortaucipir, 18F-RO948, and 18F-PI-2620, supported by a free web-based reading tool. This 5-class framework achieved high interrater and intertracer agreement, detected tau-positive individuals at earlier stages than the regulatory-approved approach, and produced tau burden classifications that showed stepwise associations with cognitive decline, Aβ load, and plasma p-tau217.

Interrater and intertracer agreement ranged from good to excellent across radiotracers, meeting or exceeding those reported for existing frameworks, while extending classification to 5 classes of tau burden.21,22,25 Composite-level agreement was notably lower for 18F-PI-2620 in composite 1 (amygdala, entorhinal cortex), where raters reported low confidence in distinguishing tau binding from off-target signal in the adjacent cavernous sinus.39 This led to the exclusion of composite 1 from 18F-PI-2620 classification criteria and explains the lower overall confidence observed for this tracer. Across tracers, 18F-MK-6240 more often yielded positive classifications when others were negative, possibly reflecting earlier tau detection,43 while 18F-PI-2620 was more frequently classified as negative, likely due to its higher positivity threshold after composite 1 exclusion. Consequently, sensitivity estimates for 18F-PI-2620 are not directly comparable to those of the other 3 tracers and should be interpreted cautiously. Despite these tracer-specific differences, overall high interrater and intertracer agreement support the feasibility of standardized reporting using a single framework.

The regulatory-approved visual framework for 18F-flortaucipir classifies scans as positive or negative, with positivity corresponding to advanced tau pathology.25,27 This approach is not sensitive to early tau deposition and does not characterize the extent of tau accumulation, limiting its ability to aid in biomarker-informed treatment decisions.3,12,32 The 5-class harmonized framework addressed both limitations. Compared with the regulatory-approved method, the harmonized framework classified significantly more participants as tau positive among CU participants (18F-flortaucipir, PR = 1.89; 18F-MK-6240, PR = 2.73) and among those with MCI (18F-flortaucipir, PR = 1.23; 18F-MK-6240, PR = 1.42), while positivity rates converged in dementia. The difference between harmonized and regulatory-approved rates widened with age, consistent with early-stage tau pathology falling below the regulatory-approved detection threshold. Participants classified as positive only by the harmonized framework had higher Aβ load and higher plasma p-tau217 levels than those negative by both frameworks, supporting the biological relevance of these additional detections.

Beyond binary classification, the 5-class tau burden classification captures the spatial progression of tau pathology in a clinically meaningful manner. Higher tau burden classes were associated with higher proportions of CI, poorer cognitive performance, higher Aβ load, and higher plasma p-tau217 in a stepwise manner across 18F-MK-6240 and 18F-flortaucipir. MoCA scores did not differ between negative and low-burden groups, consistent with early stages of AD in which tau deposition precedes measurable cognitive decline. Cognition declined progressively in subsequent stages, with reductions in MoCA scores of 2.5 to 3.3 points from low to moderate and 4.4 to 4.6 points from moderate to high burden. CLs and plasma p-tau217 increased across all burden classes, confirming that the classification tracks with both Aβ accumulation and phosphorylated tau levels.

In practice, the 5-class classification may inform diagnostic categorization, prognosis, and clinical trial eligibility more precisely than the regulatory-approved binary tau framework. A negative classification in a symptomatic individual may redirect evaluation toward non-AD etiologies.3 Among tau-positive individuals, the low and moderate burden categories identify individuals with early tau pathology, elevated Aβ burden, and absent or mild CI, representing a stage at which disease-modifying therapies may be most effective.3 Notably, some of these individuals would be classified as tau negative by the regulatory-approved framework, potentially missing an opportunity for early intervention. Individuals with high tau burden, who exhibited the greatest CI and biomarker abnormalities, carry the highest clinical risk and may have limited response to anti-Aβ therapies.32 Thus, graded tau staging may enable more refined participant selection and stratification in clinical trials than binary inclusion criteria by identifying individuals with early tau pathology who would otherwise be classified as tau negative despite representing optimal candidates for early intervention.

Limitations

This study has several limitations. Although tau PET radiotracers are used as in vivo markers of neurofibrillary tangle pathology, with strong correspondence to postmortem findings,7,8 direct neuropathological validation in this study was not performed. Primary clinical validation focused on 18F-MK-6240 and 18F-flortaucipir, the most widely used radiotracers, whereas sample sizes were smaller for 18F-PI-2620 and 18F-RO948, reflecting limited availability of participants imaged with all 4 radiotracers and potentially reducing the precision of agreement estimates. The harmonized framework was developed and evaluated in the same cohort and will require external validation in independent datasets for generalizability. The cross-sectional design precludes assessment of whether tau burden classification predicts longitudinal cognitive decline or progression across burden stages, which should be addressed in longitudinal studies. Atypical classification was uncommon in this cohort, and larger studies enriched for nonamnestic AD presentations are warranted to better characterize this category. Future work should also target better characterization of low-burden classification for 18F-PI-2620. Finally, the cohort was predominantly composed of highly educated White non-Hispanic individuals recruited from academic centers, which limits generalizability to more racially and ethnically diverse and community-based populations. Nevertheless, this study represents the largest head-to-head tau PET dataset to date, enabling direct comparison and harmonization across tracers within a single framework.

Conclusions

Per the results of this cross-sectional study, the harmonized visual reading framework for tau PET demonstrated high interrater and intertracer agreement across 4 radiotracers. Compared with the regulatory-approved framework, it detected significantly more tau-positive individuals at early disease stages. Five-class tau burden classification showed stepwise associations with cognitive performance, Aβ load, and plasma p-tau217. By enabling stratification beyond binary positivity, this harmonized tracer-agnostic framework may facilitate multicenter studies, support biomarker-informed treatment decisions and participant selection for clinical trials, and advance future clinical implementation of tau PET.

Supplement 1.

eMethods 1. Imaging Acquisition and Processing

eMethods 2. Blood Sample Measurements

eMethods 3. Harmonized Tau PET Visual Reading

eTable 1. Amyloid PET Radiotracers and Scanners per Study Site

eTable 2. Tau PET Radiotracers and Scanners per Study Site

eTable 3. Demographics and Key Characteristics From HEAD Study Participants With All Four Tau PET Radiotracers

eTable 4. Regional Tau PET Abnormality Prevalence in the Four-Tracer Subsample (n=131) and Two-Tracer Subsample (n=644)

eTable 5. Demographics and Key Characteristics From HEAD Study Participants With [18F]flortaucipir Tau PET

eTable 6. Demographics and Key Characteristics From HEAD Study Participants With [18F]RO948 Tau PET

eTable 7. Demographics and Key Characteristics From HEAD Study Participants With [18F]PI-2620 Tau PET

eTable 8. Demographics and Key Characteristics From HEAD Study per Study Site

eTable 9. Binary and Five-Level Classification Inter-Rater Agreement Across Subgroups

eTable 10. PI-2620 Inter-Rater Agreement With and Without Composite 1

eTable 11. PI-2620 Inter-Tracer Agreement With and Without Composite 1

eTable 12. Binary and Five-Level Classification Inter-Tracer Agreement Across Subgroups

eTable 13. Five-Level Classification Inter-Tracer Agreement by Scan Interval

eTable 14. PI-2620 Classification With and Without Composite 1

eTable 15. Demographics of Individuals Classified as Atypical Tau Burden Using the Harmonized Framework

eTable 16. Binary Tau PET Positivity Across Clinical Diagnoses

eTable 17. Reclassification From Regulatory-Approved to Harmonized Framework

eTable 18. Cognitive Composites (UDS-3) Across Harmonized [18F]MK-6240 Classifications

eTable 19. Cognitive Composites (UDS-3) Across Harmonized [18F]flortaucipir Classifications

eFigure 1. Study Sample Flowchart

eFigure 2. Regional Abnormality Frequencies Across Tau PET Radiotracers Informing Composite Definitions

eFigure 3. Regional Abnormality Prevalence at 1.5, 2.0, and 2.5 SD Thresholds Across Radiotracers

eFigure 4. Examples of Regions With Frequent Off-Target Binding Excluded From the Harmonized Visual Reading Framework

eFigure 5. Composite-Specific Visual Reference Masks

eFigure 6. Harmonized Tau PET Visual Reading Tool

eFigure 7. Mean Tau PET ([18F]MK-6240) SUVR Images by Harmonized Five-Level Classification

eFigure 8. Mean Tau PET ([18F]flortaucipir) SUVR Images by Harmonized Five-Level Classification

eFigure 9. Mean Tau PET ([18F]RO948) SUVR Images by Harmonized Five-Level Classification

eFigure 10. Mean Tau PET ([18F]PI-2620) SUVR Images by Harmonized Five-Level Classification

eFigure 11. Mean Tau PET SUVR Images by Harmonized Five-Level Classification Across the Four Radiotracers

eFigure 12. Inter-rater Cohen's Kappa for Tau PET Regional Composites, Five-Level Classification, and Binary Positivity

eFigure 13. Rater Confidence in Visual Reads

eFigure 14. Patterns of Inter-Tracer Disagreements

eFigure 15. Discordant Cases Where Harmonized [18F]MK-6240 Was Positive and Harmonized [18F]flortaucipir Was Negative

eFigure 16. Discordant Cases Where Harmonized [18F]MK-6240 Was Negative and Harmonized [18F]flortaucipir Was Positive

eFigure 17. Atypical Tau Burden Examples With Corresponding Radiotracer Images in Individual Participants

eFigure 18. Frequencies of 5-Level Tau Burden Classification Across the Four Tau PET Radiotracers

eFigure 19. Frequencies of Clinical Diagnoses in Each of the 5-Level Tau Burden Classification Across the Four Tau PET Radiotracers

eFigure 20. Cognitive Performance, Amyloid-β Load, and Plasma p-tau217 With Binary Tau-Positivity Across Radiotracers and Frameworks

eFigure 21. Discordant Tau PET Positivity Between Harmonized and Regulatory-Approved Visual Reading Frameworks

eReferences.

Supplement 2.

HEAD study members

Supplement 3.

Data Sharing Statement

References

  • 1.Serrano-Pozo A, Frosch MP, Masliah E, Hyman BT. Neuropathological alterations in Alzheimer disease. Cold Spring Harb Perspect Med. 2011;1(1):a006189. doi: 10.1101/cshperspect.a006189 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Jack CR Jr, Andrews JS, Beach TG, et al. Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimers Dement. 2024;20(8):5143-5169. doi: 10.1002/alz.13859 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Pascoal TA, Aguzzoli CS, Lussier FZ, et al. Insights into the use of biomarkers in clinical trials in Alzheimer’s disease. EBioMedicine. 2024;108:105322. doi: 10.1016/j.ebiom.2024.105322 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Lowe VJ, Lundt ES, Albertson SM, et al. Tau-positron emission tomography correlates with neuropathology findings. Alzheimers Dement. 2020;16(3):561-571. doi: 10.1016/j.jalz.2019.09.079 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Pascoal TA, Shin M, Kang MS, et al. In vivo quantification of neurofibrillary tangles with [18F]MK-6240. Alzheimers Res Ther. 2018;10(1):74. doi: 10.1186/s13195-018-0402-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Schöll M, Lockhart SN, Schonhaut DR, et al. PET imaging of tau deposition in the aging human brain. Neuron. 2016;89(5):971-982. doi: 10.1016/j.neuron.2016.01.028 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Aguero C, Dhaynaut M, Normandin MD, et al. Autoradiography validation of novel tau PET tracer [F-18]-MK-6240 on human postmortem brain tissue. Acta Neuropathol Commun. 2019;7(1):37. doi: 10.1186/s40478-019-0686-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Aguero C, Dhaynaut M, Amaral AC, et al. Head-to-head comparison of [18F]-flortaucipir, [18F]-MK-6240 and [18F]-PI-2620 postmortem binding across the spectrum of neurodegenerative diseases. Acta Neuropathol. 2024;147(1):25. doi: 10.1007/s00401-023-02672-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Rabinovici GD, Gatsonis C, Apgar C, et al. Association of amyloid positron emission tomography with subsequent change in clinical management among Medicare beneficiaries with mild cognitive impairment or dementia. JAMA. 2019;321(13):1286-1294. doi: 10.1001/jama.2019.2000 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Rabinovici GD, Knopman DS, Arbizu J, et al. Updated appropriate use criteria for amyloid and tau PET: a report from the Alzheimer’s Association and Society for Nuclear Medicine and Molecular Imaging Workgroup. J Nucl Med. 2025;66(suppl 2):S5-S31. doi: 10.2967/jnumed.124.268756 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Ossenkoppele R, Pichet Binette A, Groot C, et al. Amyloid and tau PET-positive cognitively unimpaired individuals are at high risk for future cognitive decline. Nat Med. 2022;28(11):2381-2387. doi: 10.1038/s41591-022-02049-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Serrano-Pozo A, Escott-Price V, Grinberg LT, et al. Alzheimer’s disease. Lancet. 2026;407(10544):2241-2262. doi: 10.1016/S0140-6736(26)00198-4 [DOI] [PubMed] [Google Scholar]
  • 13.Ossenkoppele R, Coomans EM, Apostolova LG, et al. ; Mayo Clinic Study on Aging; PREVENT-AD Research Group . Tau PET positivity in individuals with and without cognitive impairment varies with age, amyloid-β status, APOE genotype and sex. Nat Neurosci. 2025;28(8):1610-1621. doi: 10.1038/s41593-025-02000-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Moscoso A, Heeman F, Raghavan S, et al. ; PREVENT-AD Research Group, the Harvard Aging Brain Study, the LEADS Consortium, and the Alzheimer’s Disease Neuroimaging Initiative . Frequency and clinical outcomes associated with tau positron emission tomography positivity. JAMA. 2025;334(3):229-242. doi: 10.1001/jama.2025.7817 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Ossenkoppele R, Rabinovici GD, Smith R, et al. Discriminative accuracy of [18F]flortaucipir positron emission tomography for Alzheimer disease vs other neurodegenerative disorders. JAMA. 2018;320(11):1151-1162. doi: 10.1001/jama.2018.12917 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Leuzy A, Smith R, Ossenkoppele R, et al. Diagnostic performance of RO948 F 18 tau positron emission tomography in the differentiation of Alzheimer disease from other neurodegenerative disorders. JAMA Neurol. 2020;77(8):955-965. doi: 10.1001/jamaneurol.2020.0989 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Therriault J, Pascoal TA, Lussier FZ, et al. Biomarker modeling of Alzheimer’s disease using PET-based Braak staging. Nat Aging. 2022;2(6):526-535. doi: 10.1038/s43587-022-00204-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Therriault J, Schindler SE, Salvadó G, et al. Biomarker-based staging of Alzheimer disease: rationale and clinical applications. Nat Rev Neurol. 2024;20(4):232-244. doi: 10.1038/s41582-024-00942-2 [DOI] [PubMed] [Google Scholar]
  • 19.Vermeiren MR, Calandri IL, van der Flier WM, van de Giessen E, Ossenkoppele R. Survey among experts on the future role of tau-PET in clinical practice and trials. Alzheimers Dement (Amst). 2024;16(4):e70033. doi: 10.1002/dad2.70033 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Rea Reyes RE, Cody KA, Wilson RE, et al. Visual read of [F-18]florquinitau PET that includes and extends beyond the mesial temporal lobe is associated with increased plasma pTau217 and cognitive decline in a cohort that is enriched with risk for Alzheimer’s disease. Alzheimers Dement. 2025;21(2):e14406. doi: 10.1002/alz.14406 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Shuping JL, Matthews DC, Adamczuk K, et al. Development, initial validation, and application of a visual read method for [18F]MK-6240 tau PET. Alzheimers Dement (N Y). 2023;9(1):e12372. doi: 10.1002/trc2.12372 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Sonni I, Lesman Segev OH, Baker SL, et al. ; Alzheimer’s Disease Neuroimaging Initiative . Evaluation of a visual interpretation method for tau-PET with 18F-flortaucipir. Alzheimers Dement (Amst). 2020;12(1):e12133. doi: 10.1002/dad2.12133 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Seibyl JP, DuBois JM, Racine A, et al. A visual interpretation algorithm for assessing brain tauopathy with 18F-MK-6240 PET. J Nucl Med. 2023;64(3):444-451. doi: 10.2967/jnumed.122.264371 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Soleimani-Meigooni DN, Rabinovici GD. Tau PET visual reads: research and clinical applications and future directions. J Nucl Med. 2023;64(5):822-824. doi: 10.2967/jnumed.122.265017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Fleisher AS, Pontecorvo MJ, Devous MD Sr, et al. ; A16 Study Investigators . Positron emission tomography imaging with [18F]flortaucipir and postmortem assessment of Alzheimer disease neuropathologic changes. JAMA Neurol. 2020;77(7):829-839. doi: 10.1001/jamaneurol.2020.0528 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Drugs@FDA: FDA-approved drugs. US Food and Drug Administration . Accessed August 31, 2026. https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm?event=BasicSearch.process
  • 27.Drug approval package: TAUVID. US Food and Drug Administration . July 16, 2020. Accessed March 31, 2026. https://www.accessdata.fda.gov/drugsatfda_docs/nda/2020/212123Orig1s000TOC.html
  • 28.Annex I: summary of product characteristics. European Medicines Agency . Accessed March 31, 2026. https://www.ema.europa.eu/en/documents/product-information/tauvid-epar-product-information_en.pdf
  • 29.Tunali I, Wang J, Arora AK, et al. Development and validation of a 18F-flortaucipir PET visual stratification method. J Nucl Med. 2025;66(4):612-619. doi: 10.2967/jnumed.124.268700 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Hansson O. Biomarkers for neurodegenerative diseases. Nat Med. 2021;27(6):954-963. doi: 10.1038/s41591-021-01382-x [DOI] [PubMed] [Google Scholar]
  • 31.Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82(4):239-259. doi: 10.1007/BF00308809 [DOI] [PubMed] [Google Scholar]
  • 32.Sims JR, Zimmer JA, Evans CD, et al. ; TRAILBLAZER-ALZ 2 Investigators . Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA. 2023;330(6):512-527. doi: 10.1001/jama.2023.13239 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med. 2004;256(3):183-194. doi: 10.1111/j.1365-2796.2004.01388.x [DOI] [PubMed] [Google Scholar]
  • 34.Klunk WE, Koeppe RA, Price JC, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dement. 2015;11(1):1-15.e1, 4. doi: 10.1016/j.jalz.2014.07.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Manera AL, Dadar M, Fonov V, Collins DL. CerebrA, registration and manual label correction of Mindboggle-101 atlas for MNI-ICBM152 template. Sci Data. 2020;7(1):237. doi: 10.1038/s41597-020-0557-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Ikonomovic MD, Abrahamson EE, Price JC, Mathis CA, Klunk WE. [F-18]AV-1451 positron emission tomography retention in choroid plexus: more than “off-target” binding. Ann Neurol. 2016;80(2):307-308. doi: 10.1002/ana.24706 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Baker SL, Harrison TM, Maass A, La Joie R, Jagust WJ. Effect of off-target binding on 18F-flortaucipir variability in healthy controls across the life span. J Nucl Med. 2019;60(10):1444-1451. doi: 10.2967/jnumed.118.224113 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Tissot C, Servaes S, Lussier FZ, et al. The association of age-related and off-target retention with longitudinal quantification of [18F]MK6240 tau PET in target regions. J Nucl Med. 2023;64(3):452-459. doi: 10.2967/jnumed.122.264434 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Tonietto M, Sotolongo-Grau O, Roé-Vellvé N, et al. ; FACEHBI study group; AMYPAD consortium . Head-to-head comparison of tau PET tracers [18F]PI-2620 and [18F]RO948 in non-demented individuals with brain amyloid deposition: the TAU-PET FACEHBI cohort. Alzheimers Res Ther. 2024;16(1):257. doi: 10.1186/s13195-024-01622-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Fedorov A, Beichel R, Kalpathy-Cramer J, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging. 2012;30(9):1323-1341. doi: 10.1016/j.mri.2012.05.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Cicchetti DV, Sparrow SA. Developing criteria for establishing interrater reliability of specific items: applications to assessment of adaptive behavior. Am J Ment Defic. 1981;86(2):127-137. [PubMed] [Google Scholar]
  • 42.Weintraub S, Besser L, Dodge HH, et al. Version 3 of the Alzheimer Disease Centers’ Neuropsychological Test Battery in the Uniform Data Set (UDS). Alzheimer Dis Assoc Disord. 2018;32(1):10-17. doi: 10.1097/WAD.0000000000000223 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Povala G, Bellaver B, Lussier FZ, et al. HEAD Study. Comparison of [18F]flortaucipir and [18F]MK6240 for the detection of tau pathology in Alzheimer’s disease (HEAD): a multicentre, prospective, cross-sectional, within-participant study. Lancet. 2026;407(10544):2217-2226. doi: 10.1016/S0140-6736(26)00417-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

eMethods 1. Imaging Acquisition and Processing

eMethods 2. Blood Sample Measurements

eMethods 3. Harmonized Tau PET Visual Reading

eTable 1. Amyloid PET Radiotracers and Scanners per Study Site

eTable 2. Tau PET Radiotracers and Scanners per Study Site

eTable 3. Demographics and Key Characteristics From HEAD Study Participants With All Four Tau PET Radiotracers

eTable 4. Regional Tau PET Abnormality Prevalence in the Four-Tracer Subsample (n=131) and Two-Tracer Subsample (n=644)

eTable 5. Demographics and Key Characteristics From HEAD Study Participants With [18F]flortaucipir Tau PET

eTable 6. Demographics and Key Characteristics From HEAD Study Participants With [18F]RO948 Tau PET

eTable 7. Demographics and Key Characteristics From HEAD Study Participants With [18F]PI-2620 Tau PET

eTable 8. Demographics and Key Characteristics From HEAD Study per Study Site

eTable 9. Binary and Five-Level Classification Inter-Rater Agreement Across Subgroups

eTable 10. PI-2620 Inter-Rater Agreement With and Without Composite 1

eTable 11. PI-2620 Inter-Tracer Agreement With and Without Composite 1

eTable 12. Binary and Five-Level Classification Inter-Tracer Agreement Across Subgroups

eTable 13. Five-Level Classification Inter-Tracer Agreement by Scan Interval

eTable 14. PI-2620 Classification With and Without Composite 1

eTable 15. Demographics of Individuals Classified as Atypical Tau Burden Using the Harmonized Framework

eTable 16. Binary Tau PET Positivity Across Clinical Diagnoses

eTable 17. Reclassification From Regulatory-Approved to Harmonized Framework

eTable 18. Cognitive Composites (UDS-3) Across Harmonized [18F]MK-6240 Classifications

eTable 19. Cognitive Composites (UDS-3) Across Harmonized [18F]flortaucipir Classifications

eFigure 1. Study Sample Flowchart

eFigure 2. Regional Abnormality Frequencies Across Tau PET Radiotracers Informing Composite Definitions

eFigure 3. Regional Abnormality Prevalence at 1.5, 2.0, and 2.5 SD Thresholds Across Radiotracers

eFigure 4. Examples of Regions With Frequent Off-Target Binding Excluded From the Harmonized Visual Reading Framework

eFigure 5. Composite-Specific Visual Reference Masks

eFigure 6. Harmonized Tau PET Visual Reading Tool

eFigure 7. Mean Tau PET ([18F]MK-6240) SUVR Images by Harmonized Five-Level Classification

eFigure 8. Mean Tau PET ([18F]flortaucipir) SUVR Images by Harmonized Five-Level Classification

eFigure 9. Mean Tau PET ([18F]RO948) SUVR Images by Harmonized Five-Level Classification

eFigure 10. Mean Tau PET ([18F]PI-2620) SUVR Images by Harmonized Five-Level Classification

eFigure 11. Mean Tau PET SUVR Images by Harmonized Five-Level Classification Across the Four Radiotracers

eFigure 12. Inter-rater Cohen's Kappa for Tau PET Regional Composites, Five-Level Classification, and Binary Positivity

eFigure 13. Rater Confidence in Visual Reads

eFigure 14. Patterns of Inter-Tracer Disagreements

eFigure 15. Discordant Cases Where Harmonized [18F]MK-6240 Was Positive and Harmonized [18F]flortaucipir Was Negative

eFigure 16. Discordant Cases Where Harmonized [18F]MK-6240 Was Negative and Harmonized [18F]flortaucipir Was Positive

eFigure 17. Atypical Tau Burden Examples With Corresponding Radiotracer Images in Individual Participants

eFigure 18. Frequencies of 5-Level Tau Burden Classification Across the Four Tau PET Radiotracers

eFigure 19. Frequencies of Clinical Diagnoses in Each of the 5-Level Tau Burden Classification Across the Four Tau PET Radiotracers

eFigure 20. Cognitive Performance, Amyloid-β Load, and Plasma p-tau217 With Binary Tau-Positivity Across Radiotracers and Frameworks

eFigure 21. Discordant Tau PET Positivity Between Harmonized and Regulatory-Approved Visual Reading Frameworks

eReferences.

Supplement 2.

HEAD study members

Supplement 3.

Data Sharing Statement


Articles from JAMA Neurology are provided here courtesy of American Medical Association

RESOURCES